Because of one user's misconfiguration of Microsoft's HypeV, his virtual machines were not getting the results of arp. As a result of that configuration all the packets going to machines on the same subnetwork were going to the default gateway. The default gateway was an OpenBSD 6.1 server. OpenBSD very slowly forward the packets back out the same if (an em0) and the packets got to where they were supposed to go.
I a long time ago I tried to redirect, using pf, an external ip address back to an internal ip address. It did not work, and I believe I was told it could not work. So I am surprised that the above was working at all. I also don't understand why it was so slow. The existence of the problem was discovered when the transmission time were so slow the timeout occurred.
